workflow.failOnIgnore not reflected in workflow.success / workflow.exitStatus (onComplete + reporting see success)

[adamrtalbot]

Bug report

Summary

When workflow.failOnIgnore = true and a task fails under errorStrategy 'ignore', the nextflow process correctly exits with code 1, but workflow.success is still true and workflow.exitStatus is 0.

Because workflow.success / workflow.exitStatus are what the workflow.onComplete handler, the built-in email notification, and downstream trace/reporting integrations consume, those all report the run as successful even though the CLI exit code is a failure. Anything that keys off workflow.onComplete (dashboards, notifications, monitoring) sees a green run.

This is the same class of bug as #6962 (AgentLogObserver printing [SUCCESS] on a failed run), which was fixed by switching that observer to session.isSuccess(). WorkflowMetadata.success was not updated and still ignores failOnIgnore. It's also the underlying reason behind the confusion in #6297 — the reporter there only ever checked the exit code, so "upgrading fixed it" masked the fact that workflow.success is still wrong.

Minimal reproducible example

main.nf:

process FAILME {
    errorStrategy 'ignore'
    script:
    """
    echo boom; exit 1
    """
}

workflow {
    FAILME()
}

nextflow.config:

workflow {
    failOnIgnore = true
    onComplete = {
        println "onComplete -> workflow.success=${workflow.success} workflow.exitStatus=${workflow.exitStatus}"
    }
}

Run and capture the process exit code:

nextflow run main.nf ; echo "nextflow exit code = $?"

Actual behaviour

onComplete -> workflow.success=true workflow.exitStatus=0
nextflow exit code = 1

workflow.success is true and workflow.exitStatus is 0, contradicting the process exit code of 1.

Expected behaviour

When failOnIgnore triggers a failure, workflow.success should be false and workflow.exitStatus should be non-zero, consistent with the CLI exit code, so onComplete/email/reporting see the run as failed.

Root cause

Two separate success computations disagree:

• Session.isSuccess() — !aborted && !cancelled && !failOnIgnore — drives the CLI exit code (via ScriptRunner). Correctly returns failure.
• WorkflowMetadata.invokeOnComplete() sets this.success = !(session.aborted || session.cancelled), which omits failOnIgnore. Since failOnIgnore sets no session.fault, setErrorAttributes() falls through to the else branch and assigns exitStatus = 0.

Suggested fix (mirrors #6962): base WorkflowMetadata.success on session.isSuccess() and set a non-zero exitStatus on the failOnIgnore path.

Environment

• Nextflow version: 26.04.4 (also reproducible on the 25.04 line)
• Java: 17
• OS: macOS
• Executor: local (not executor-specific — the divergence is in WorkflowMetadata)

[ehsanestaji]

Confirmed locally on Nextflow 26.04.4 with Java 25: workflow.success=true and workflow.exitStatus=0, while the Nextflow process exits with status 1. I’d like to add regression coverage in WorkflowMetadataTest and work on the fix. For the failOnIgnore path, should metadata use a generic exit status of 1, or preserve the ignored task’s exit status?

[adamrtalbot]

For the failOnIgnore path, should metadata use a generic exit status of 1, or preserve the ignored task’s exit status?

exit 1, nextflow is exiting with an error, not the task.

Nextflow could also exit with it's own error, e.g. 137 out of memory and it's important to distinguish these.